Each layer is tried in order. The system only escalates to the next (more expensive) layer when the current one can't achieve the required confidence threshold. This minimizes browser renders and maximizes throughput.
HTTP + Regex + Schema.org
Fast HTTP requests with BeautifulSoup parsing. Extracts emails, phone numbers, and structured data (JSON-LD, schema.org microdata) from raw HTML. Handles 60-70% of company websites that serve static content.
Escalates if confidence < threshold or page returns empty/minimal content
__NEXT_DATA__, React Props, API Endpoints
Extracts data from JavaScript-rendered frameworks without launching a browser. Parses __NEXT_DATA__ (Next.js), data-react-props attributes, inline JSON payloads, and discovers internal API endpoints that serve structured data.
Escalates if no JS data structures found or extracted data is incomplete
Page Classification & Priority Crawling
Classifies discovered pages by type (/team, /about, /contact, /leadership, /staff) and crawls them in priority order. Budget-aware: max 15 pages per company with intelligent page selection.
Escalates if classified pages don't contain extractable contact data
4-Method Contact Extraction
Deep content analysis using 4 cascading extraction methods. Combines structured data parsing, visual layout analysis, proximity heuristics, and social profile matching to find decision makers.
Escalates if semantic methods find < 2 contacts with confidence > 0.5
Playwright Stealth Rendering
Full Playwright browser rendering for JavaScript-heavy SPAs that resist all other methods. Budget-capped at 3 browser renders per company to control costs. Handles click-to-reveal content, infinite scroll, and modal-based team directories.
Final layer -- if browser rendering fails, the company is marked for manual review
Everything you need to know about our platform.
Still have questions?
Our team can walk you through the pipeline, pricing, and your use case.